Politics of AI
Posts
Colorado's Example

Colorado's Example

Daniel Dean
June 29, 2024

This week's edition of Politics of AI features an analysis of Colorado's recent AI legislation, yet another deepfake, and culminates with a SPACE POTATO!

Colorado’s AI Legislation

Colorado's AI Act, SB 24-205, regulates AI use, targeting developers and deployers to prevent algorithmic discrimination.

What’s new? The legislation is the first comprehensive AI legislation in the country, going further than that of any other state. Developers must disclose system risks, provide documentation, and notify the state Attorney General of algorithmic discrimination within 90 days. Deployers must complete impact assessments, notify consumers, and disclose discrimination findings to the attorney general.

Who deals with it? Enforcement is solely by the state Attorney General; there's no private right of action under the law.

What’s the problem? Enforcement is vague in Colorado’s legislation. So far, it would appear the only way to enforce any pieces of the legislation would be in the court system and individual victims of AI malfeasance have no recourse. Between now and the legislation’s effective date, look for the office of Colorado Attorney General Philip J. Weiser (D) to issue guidance and rules on the matter over the next year and a half.

How can they know? Without some kind of registration system, how can an Attorney Generals office know which AI systems are based in their state? When an AI system is in operation, it is not obvious that it’s there to outside parties. Colorado could have some AI systems harming its citizens and have no idea anything is happening to begin with. With AI’s capabilities still being fairly limited, this may not yet be a major problem, but as AI systems advance and, more importantly, become cheaper to create, tracking them will become difficult.

When does it start? Starting February 1, 2026, developers and deployers of high-risk AI systems must use reasonable care to protect consumers from known or foreseeable algorithmic discrimination risks.